Robotic Manipulation on LIBERO (Success Rates and Efficiency)
99.1Long Success RateAdaWAM
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| AdaWAM2026.06 | 99.1 | 98 | 99.6 | 97.1 | 98.5 | — | — | — | — | |
| LingBot-VAPrediction Mode=video-action joint2026.06 | 98.5 | 98.5 | 99.6 | 97.2 | 98.5 | — | — | — | — | |
| π0.5 with MoHParadigm=Flow-matching or diffusion-based VLA, Size=3B, Iters=30k2025.11 | 98.4 | 98.8 | 100 | 98.8 | 99 | — | — | — | — | |
| X-VLAParadigm=Flow-matching or diffusion-based VLA, Size=1B, Iters=60k†2025.11 | 97.6 | 98.2 | 98.6 | 97.8 | 98.1 | — | — | — | — | |
| X-VLA2026.06 | 97.6 | 98.2 | 98.6 | 97.8 | 98.1 | — | — | — | — | |
| MotusPrediction Mode=video-action joint2026.06 | 97.6 | 96.8 | 99.8 | 96.6 | 97.7 | — | — | — | — | |
| AdaWAM w/o. T.R.Reasoning Module=Textual reasoning removed2026.06 | 97.4 | — | — | — | — | — | — | — | — | |
| ACoT-VLA2026.06 | 97 | 98.6 | 99 | 99.4 | 98.5 | — | — | — | — | |
| AdaWAM w/o. V.R.Reasoning Module=Visual reasoning removed2026.06 | 96.6 | 97.5 | 99.4 | 96.8 | 97.6 | — | — | — | — | |
| DRIFTBase Model=Qwen3-VL-2B [2], Action Decoder=DRIFT-MLP2026.06 | 96.4 | 98.8 | 99.2 | 97 | 97.9 | — | — | — | — | |
| Spatial ForcingParadigm=Flow-matching or diffusion-based VLA, Size=7B, Iters=150k2025.11 | 96 | 99.4 | 99.6 | 98.8 | 98.5 | — | — | — | — | |
| π0.5Paradigm=Flow-matching or diffusion-based VLA, Size=3B, Iters=30k2025.11 | 95.4 | 98.8 | 99 | 97.6 | 97.7 | — | — | — | — | |
| Fast-WAMPrediction Mode=action-only2026.06 | 95.2 | 98.2 | 100 | 97 | 97.6 | — | — | — | — | |
| VLA-AdapterParadigm=Flow-matching or diffusion-based VLA, Size=0.5B, Iters=150k2025.11 | 95 | 97.8 | 99.2 | 97.2 | 97.3 | — | — | — | — | |
| VLA-AdaptorBase Model=Qwen-2.5-0.5B [60], Action Decoder=Diffusion2026.06 | 95 | 97.8 | 99.2 | 97.2 | 97.3 | — | — | — | — | |
| Coarse-to-ControlCoT Type=Action CoT2026.06 | 95 | 98.8 | 100 | 97.8 | 97.9 | — | — | — | — | |
| OpenVLA-OFTParadigm=Flow-matching or diffusion-based VLA, Size=7B, Iters=150k2025.11 | 94.5 | 97.6 | 98.4 | 97.9 | 97.1 | — | — | — | — | |
| OpenVLA-OFTCoT Type=No CoT2026.06 | 94.5 | 97.6 | 98.4 | 97.9 | 97.1 | — | — | — | — | |
| OpenVLA-OFTAction Pretraining (Pt.)=Yes2026.06 | 94.5 | 97.6 | 98.4 | 97.9 | 97.1 | — | — | — | — | |
| GR00T-N1.6Action Pretraining (Pt.)=Yes2026.06 | 94.4 | 97.7 | 98.5 | 97.5 | 97 | — | — | — | — | |
| UniviewVLA2026.06 | 94.2 | 96.4 | 99.2 | 93.4 | 95.8 | — | — | — | — | |
| UniVLACoT Type=Visual CoT2026.06 | 94 | 95.4 | 98.8 | 93.6 | 95.5 | — | — | — | — | |
| Embodied-R1.5-VLAAction Pretraining (Pt.)=No2026.06 | 93.9 | 98.3 | 99.2 | 97.6 | 97.3 | — | — | — | — | |
| VLA-PrunerBase Model=OpenVLA-OFT, Token Retention Ratio=50%2025.11 | 92.6 | 97.3 | 98.6 | 96.8 | — | 101.05 | 2.234 | 92.86 | 1.46 | |
| InternVLABase Model=Qwen2.5-VL-3B [3], Action Decoder=Diffusion2026.06 | 92.6 | 98 | 99 | 93.8 | 95.8 | — | — | — | — | |
| StarVLA with MoHParadigm=Flow-matching or diffusion-based VLA, Size=3B, Iters=30k2025.11 | 92.4 | 98.4 | 99.6 | 97.6 | 97 | — | — | — | — | |
| π0.5Base Model=PaliGemma-3B [4], Action Decoder=Flow2026.06 | 92.4 | 98.8 | 98.2 | 98 | 96.9 | — | — | — | — | |
| π0.5CoT Type=Textual CoT2026.06 | 92.4 | 98.8 | 98.2 | 98 | 96.8 | — | — | — | — | |
| π0.5Action Pretraining (Pt.)=Yes2026.06 | 92.4 | 98.8 | 98.2 | 98 | 96.9 | — | — | — | — | |
| UniVLAParadigm=Regression or classification-based VLA, Size=9B, Iters=8k†2025.11 | 92 | 96.5 | 96.8 | 95.6 | 95.2 | — | — | — | — | |
| πreg with MoHParadigm=Regression or classification-based VLA, Size=3B, Iters=30k2025.11 | 91.4 | 99 | 98.8 | 96.4 | 96.4 | — | — | — | — | |
| StarVLAParadigm=Flow-matching or diffusion-based VLA, Size=3B, Iters=30k2025.11 | 91.4 | 98 | 98.2 | 95.8 | 95.9 | — | — | — | — | |
| UniVLA2026.06 | 91.4 | 96 | 99.2 | 93.2 | 95 | — | — | — | — | |
| F1CoT Type=Visual CoT2026.06 | 91.3 | 98.2 | 97.8 | 95.4 | 95.7 | — | — | — | — | |
| VanillaBase Model=OpenVLA-OFT, Token Retention Ratio=100%2025.11 | 90.7 | 95.8 | 98.7 | 96.3 | — | 100 | 3.903 | 135.78 | 1 | |
| OpenVLA-OFT2026.06 | 90.7 | 96.2 | 98.3 | 96.2 | 95.4 | — | — | — | — | |
| GR00T-N1Paradigm=Flow-matching or diffusion-based VLA, Size=3B, Iters=100k2025.11 | 90.6 | 94.4 | 97.6 | 93 | 93.9 | — | — | — | — | |
| GR00T-N1CoT Type=No CoT2026.06 | 90.6 | 94.4 | 97.6 | 93 | 93.9 | — | — | — | — | |
| GR00T-N1Action Pretraining (Pt.)=Yes2026.06 | 90.6 | 94.4 | 97.6 | 93 | 93.9 | — | — | — | — | |
| GR00T N12026.06 | 90.6 | 94.4 | 97.6 | 93 | 93.9 | — | — | — | — | |
| VLA-CacheBase Model=OpenVLA-OFT, Token Retention Ratio=50%2025.11 | 90.2 | 95.4 | 96 | 96.7 | — | 99.09 | 2.73 | 101.01 | 1.34 | |
| VLA-PrunerBase Model=OpenVLA-OFT, Token Retention Ratio=25%2025.11 | 90.2 | 93.5 | 96.2 | 95.2 | — | 98.37 | 1.42 | 75.64 | 1.8 | |
| πregParadigm=Regression or classification-based VLA, Size=3B, Iters=30k2025.11 | 90.2 | 97.8 | 98.2 | 94.6 | 95.2 | — | — | — | — | |
| UD-VLACoT Type=Visual CoT2026.06 | 89.6 | 94.1 | 95.7 | 91.2 | 92.7 | — | — | — | — | |
| DreamVLACoT Type=Visual CoT2026.06 | 89.5 | 97.5 | 94 | 89.5 | 92.6 | — | — | — | — | |
| MM-ACT2026.06 | 88 | 97.8 | 99.4 | 94.8 | 95 | — | — | — | — | |
| VLA-02026.06 | 87.6 | 97 | 97.8 | 96.2 | 94.7 | — | — | — | — | |
| π0 with MoHParadigm=Flow-matching or diffusion-based VLA, Size=3B, Iters=30k2025.11 | 87.4 | 97.6 | 98.8 | 96.4 | 95.1 | — | — | — | — | |
| FastVBase Model=OpenVLA-OFT, Token Retention Ratio=50%2025.11 | 87 | 94.6 | 96.8 | 92.7 | — | 97.26 | 2.219 | 88.23 | 1.54 | |
| OpenVLA-OFTAction Pretraining (Pt.)=No2026.06 | 86.5 | 94.3 | 95.2 | 91.7 | 91.9 | — | — | — | — | |
| π0.5Action Pretraining (Pt.)=No2026.06 | 85.8 | 96.6 | 97.2 | 94.6 | 93.6 | — | — | — | — | |
| π0Base Model=PaliGemma-3B [4], Action Decoder=Flow2026.06 | 85.2 | 96.8 | 98.8 | 95.8 | 94.1 | — | — | — | — | |
| π02026.06 | 85.2 | 96.8 | 98.8 | 95.8 | 94.2 | — | — | — | — | |
| π0CoT Type=No CoT2026.06 | 85.2 | 96.8 | 98.8 | 95.8 | 94.2 | — | — | — | — | |
| π0Action Pretraining (Pt.)=Yes2026.06 | 85.2 | 96.8 | 98.8 | 95.8 | 94.2 | — | — | — | — | |
| SparseVLMBase Model=OpenVLA-OFT, Token Retention Ratio=50%2025.11 | 85.1 | 94.1 | 93.7 | 91.2 | — | 95.37 | 2.289 | 90.22 | 1.5 | |
| π0Paradigm=Flow-matching or diffusion-based VLA, Size=3B, Iters=30k2025.11 | 84.2 | 97.4 | 98.2 | 95.4 | 93.8 | — | — | — | — | |
| DivPruneBase Model=OpenVLA-OFT, Token Retention Ratio=50%2025.11 | 83.1 | 90.8 | 91.1 | 89.9 | — | 92.61 | 2.126 | 88.01 | 1.54 | |
| VLA-CacheBase Model=OpenVLA-OFT, Token Retention Ratio=25%2025.11 | 80.3 | 85.6 | 85.3 | 84.1 | — | 88.11 | 1.938 | 85.02 | 1.6 | |
| SparseVLMBase Model=OpenVLA-OFT, Token Retention Ratio=25%2025.11 | 79.1 | 89.8 | 84.8 | 82.7 | — | 88.63 | 1.459 | 74.71 | 1.82 | |
| MolmoAct-7B-DCoT Type=Action CoT2026.06 | 77.2 | 87 | 95.4 | 87.6 | 86.6 | — | — | — | — | |
| SmolVLAParadigm=Flow-matching or diffusion-based VLA, Size=2B, Iters=100k2025.11 | 77 | 93 | 94 | 91 | 88.8 | — | — | — | — | |
| SmolVLACoT Type=No CoT2026.06 | 77 | 93 | 94 | 91 | 88.8 | — | — | — | — | |
| FastVBase Model=OpenVLA-OFT, Token Retention Ratio=25%2025.11 | 74.6 | 87.8 | 81.8 | 87.6 | — | 87.48 | 1.401 | 72.73 | 1.87 | |
| DivPruneBase Model=OpenVLA-OFT, Token Retention Ratio=25%2025.11 | 72.5 | 83.9 | 72.6 | 79.3 | — | 80.76 | 1.389 | 72.3 | 1.88 | |
| ThinkActCoT Type=Textual CoT2026.06 | 70.9 | 88.3 | 91.4 | 87.1 | 84.4 | — | — | — | — | |
| CoT-VLAParadigm=Regression or classification-based VLA, Size=7B, Iters=100k2025.11 | 69 | 87.5 | 91.6 | 87.6 | 83.9 | — | — | — | — | |
| CoT-VLA2026.06 | 69 | 87.5 | 91.6 | 87.6 | 81.1 | — | — | — | — | |
| CoT-VLA-7BCoT Type=Visual CoT2026.06 | 69 | 87.5 | 91.6 | 87.6 | 81.1 | — | — | — | — | |
| CoT-VLA2026.06 | 69 | 87.5 | 91.6 | 87.6 | 83.9 | — | — | — | — | |
| VLA-PrunerBase Model=OpenVLA-OFT, Token Retention Ratio=12.5%2025.11 | 68.8 | 88.1 | 87.6 | 84.9 | — | 88.27 | 1.096 | 68.95 | 1.99 | |
| VLM4VLABase Model=InternVL3.5-4B [55], Action Decoder=MLP2026.06 | 62.8 | — | — | — | — | — | — | — | — | |
| π0-FASTParadigm=Regression or classification-based VLA, Size=3B, Iters=30k2025.11 | 60.2 | 96.4 | 96.8 | 88.6 | 85.5 | — | — | — | — | |
| π0-FASTCoT Type=No CoT2026.06 | 60.2 | 96.4 | 96.8 | 88.6 | 85.5 | — | — | — | — | |
| π0-FASTAction Pretraining (Pt.)=Yes2026.06 | 60.2 | 96.4 | 96.8 | 88.6 | 85.5 | — | — | — | — | |
| π0-FAST2026.06 | 60.2 | 96.4 | 96.8 | 88.6 | 85.5 | — | — | — | — | |
| WorldVLACoT Type=Visual CoT2026.06 | 60 | 87.6 | 96.2 | 83.4 | 81.8 | — | — | — | — | |
| DRIFTBase Model=OpenVLA-7B [20], Action Decoder=DRIFT-Tokenizer2026.06 | 57 | 87 | 89.1 | 77.5 | 77.7 | — | — | — | — | |
| VLA-PrunerBase Model=OpenVLA, Token Retention Ratio=50%2025.11 | 56.4 | 88.2 | 85.8 | 79.4 | — | 102.45 | 1.139 | 178.38 | 1.33 | |
| VLM4VLABase Model=Qwen3-VL-2B [2], Action Decoder=MLP2026.06 | 55.8 | — | — | — | — | — | — | — | — | |
| OpenVLA*Base Model=Prismatic-7B [17], Action Decoder=Tokenizer2026.06 | 54.1 | 83.1 | 88.5 | 76.4 | 75.5 | — | — | — | — | |
| OpenVLAParadigm=Regression or classification-based VLA, Size=7B, Iters=150k2025.11 | 53.7 | 84.7 | 88.4 | 79.2 | 76.5 | — | — | — | — | |
| OpenVLA2026.06 | 53.7 | 84.7 | 88.4 | 79.2 | 76.5 | — | — | — | — | |
| OpenVLAAction Pretraining (Pt.)=Yes2026.06 | 53.7 | 84.7 | 88.4 | 79.2 | 76.5 | — | — | — | — | |
| OpenVLA2026.06 | 53.7 | 84.9 | 88.4 | 79.2 | 76.6 | — | — | — | — | |
| VanillaBase Model=OpenVLA, Token Retention Ratio=100%2025.11 | 52.2 | 87.6 | 84.6 | 78.6 | — | 100 | 1.906 | 236.41 | 1 | |
| VLA-CacheBase Model=OpenVLA, Token Retention Ratio=50%2025.11 | 51.8 | 87.1 | 80.7 | 78.6 | — | 98.48 | 1.384 | 192.2 | 1.23 | |
| VLA-PrunerBase Model=OpenVLA, Token Retention Ratio=25%2025.11 | 51.8 | 85.4 | 82.5 | 78.4 | — | 98.48 | 0.758 | 144.81 | 1.63 | |
| OctoParadigm=Regression or classification-based VLA, Size=0.1B, Iters=-2025.11 | 51.1 | 78.9 | 85.7 | 84.6 | 75.1 | — | — | — | — | |
| FastVBase Model=OpenVLA, Token Retention Ratio=50%2025.11 | 50.6 | 86.2 | 81.6 | 77.2 | — | 97.43 | 1.136 | 172.32 | 1.37 | |
| Diffusion PolicyParadigm=Flow-matching or diffusion-based VLA, Size=30M, Iters=-2025.11 | 50.5 | 78.3 | 92.5 | 68.3 | 72.4 | — | — | — | — | |
| Diffusion PolicyAction Pretraining (Pt.)=No2026.06 | 50.5 | 78.3 | 92.5 | 68.3 | 72.4 | — | — | — | — | |
| SparseVLMBase Model=OpenVLA, Token Retention Ratio=50%2025.11 | 48.6 | 85.6 | 80.5 | 75 | — | 95.44 | 1.155 | 175.77 | 1.34 | |
| π0-FASTAction Pretraining (Pt.)=No2026.06 | 48 | 87 | 63 | 89 | 71.8 | — | — | — | — | |
| DivPruneBase Model=OpenVLA, Token Retention Ratio=50%2025.11 | 47.6 | 82.6 | 78.8 | 71.8 | — | 92.64 | 1.105 | 173.88 | 1.36 | |
| SparseVLMBase Model=OpenVLA, Token Retention Ratio=25%2025.11 | 45.6 | 83.4 | 72.8 | 67.6 | — | 88.67 | 0.772 | 144.12 | 1.64 | |
| VLA-CacheBase Model=OpenVLA, Token Retention Ratio=25%2025.11 | 45.5 | 78.1 | 73.2 | 70.2 | — | 88.08 | 0.961 | 164.17 | 1.44 | |
| FastVBase Model=OpenVLA, Token Retention Ratio=25%2025.11 | 43.8 | 81.6 | 69.6 | 71.6 | — | 87.62 | 0.757 | 141.25 | 1.67 | |
| VLA-PrunerBase Model=OpenVLA, Token Retention Ratio=12.5%2025.11 | 42.8 | 80.2 | 78.4 | 69 | — | 88.9 | 0.571 | 129.01 | 1.83 | |
| DivPruneBase Model=OpenVLA, Token Retention Ratio=25%2025.11 | 41.4 | 77.4 | 61.3 | 65.4 | — | 80.96 | 0.743 | 142.12 | 1.66 |